Neural Network-Based Models for Machine Translation Systems (EN > {FR,SL,ES,DE,IT};{FR,SL,ES,DE,IT,NL} > EN;FR <> DE;ES <> PT)

These are models for machine translation between English (EN), French (FR), Slovenian (SL), Spanish (ES), German (DE), Italian (IT) and Dutch (NL). In particular, a model is included for each of the following language pairs (in “source language > target language” format): EN>FR, EN>SL, EN>ES, EN>DE, EN>IT, FR>EN, SL>EN, ES>EN, DE>EN, IT>EN, NL>EN, FR>DE, DE>FR, ES>PT and PT>ES. They solve the problem of automatic translation of documents, closed and real-time subtitles, and simultaneous conversations. These are cutting-edge models, similar to those used by the best current commercial systems of large technology companies. They are based on large amounts of data from different sources (educational, television, journalistic, legal …), which allows obtaining high precision results in different areas. Our systems based on these models have obtained first places in prestigious international competitions (RTVE-IberSpeech, TV Speech-to-Text Challenge; International Conference on Machine Translation, WMT). In addition, the MLLP-VRAIN group of the UPV can customize the models adapting them to the client’s domain to increase the accuracy of the results. From a technical point of view, these models are trained deep neural networks. The automatic translation software (registered in UPV, ref. S-19912-2018) requires one of these models to perform automatic translations between the languages of the given model.